v3.15.0

What's Changed

• release: sync Formula sha256 to v3.14.0 tarball by @MongLong0214 in #663
• test(#289): probe the stale-write counter, and record what it cannot prove by @MongLong0214 in #665
• test(#289): pin the safety properties the compare-and-swap already provides by @MongLong0214 in #666
• test(#289): pin the postPoll coupling — a slow subscriber holds the poller by @MongLong0214 in #667
• test(#289): pin the subscriber coupling link by link, with no calibrated durations by @MongLong0214 in #671
• test: give the AX-backed probe a budget it can actually observe the answer in by @MongLong0214 in #672
• test: stop the qualification summary from stating a count it never counted by @MongLong0214 in #673
• fix(#668): the refresh receipt reports writes that landed, not reads that decoded by @MongLong0214 in #674
• fix(#668): coalesce concurrent refresh cycles, serving late callers a fresh one by @MongLong0214 in #675
• probe(#668): make the one measurement this issue turns on executable by @MongLong0214 in #676
• chore(#289): retire ADR-006's unadopted snapshot capsule, keep the part in use by @MongLong0214 in #677
• docs(#678): promote docs/roadmap/ to source of truth, retire the out-of-repo structure by @MongLong0214 in #679
• feat(#373): a Phase-B readback must prove it postdates the mutation it witnesses by @MongLong0214 in #680
• feat(#373): print which read-only operations are short of passed, and correct the note that guessed by @MongLong0214 in #681
• fix(#373): an unavailable deferral says what the operation said, not what the runner guessed by @MongLong0214 in #682
• ci(#678): fail when the roadmap table and GitHub disagree by @MongLong0214 in #684
• fix(#678): let a closing pull request mark its own issue closed by @MongLong0214 in #686
• fix(#290): identify the track-header pan slider instead of eliminating to it by @MongLong0214 in #687
• feat(livekit): make a check prove it can reject the state it names by @MongLong0214 in #688
• chore(livekit): drop the mutation_flips / mutation_backed aliases by @MongLong0214 in #689
• feat(#290): measure the three track-header toggle locators by @MongLong0214 in #690
• fix(#290): the transport bar narrows instead of taking the first in tree order by @MongLong0214 in #691
• fix(#685): a failed AX read no longer abandons the write partway by @MongLong0214 in #692
• fix(#300): a band that measured nothing no longer reports silence by @MongLong0214 in #693
• fix(#300): an accumulator band is no longer picked as a resonance by @MongLong0214 in #694
• feat(#300): promote the spectral commands by @MongLong0214 in #695
• fix(#300): an accumulator band is not a peak either by @MongLong0214 in #696
• feat(#300): recommendations carry a confidence derived from what decided them by @MongLong0214 in #697
• docs: mark #300 closed by @MongLong0214 in #698
• test(#290): the atlas scoring cannot resolve the elements Logic exposes by @MongLong0214 in #699
• feat(#290): score selectors as a ratio over the evidence they asked for by @MongLong0214 in #700
• feat(#290): adopt the atlas at its first production call site by @MongLong0214 in #701
• feat(#290): adopt the atlas for the volume fader, removing two positional paths by @MongLong0214 in #702
• refactor(#290): give the atlas the product's match modes, and adopt the toggles by @MongLong0214 in #703
• feat(#290): adopt the atlas for the control bar, and name what it cannot express by @MongLong0214 in #704
• fix(#290): give the adopted selectors their own identities by @MongLong0214 in #705
• feat(#290): capture redacted AX baselines, and one Desktop/ko fixture by @MongLong0214 in #706
• fix(#290): diff two atlas baselines, and close three fail-open paths by @MongLong0214 in #707
• fix(#290): cover the control bar, and the thing transport actually uses by @MongLong0214 in #708
• docs(#290): the roadmap said the atlas has zero consumers, and it has six by @MongLong0214 in #709
• feat: a Sources change now obliges the harness that claims it by @MongLong0214 in #710
• fix(#293): the Event List readback could not run in any language but English by @MongLong0214 in #712
• docs: two roadmap rows described walls that measurement does not find by @MongLong0214 in #713
• fix(#293): the run's own tidying was inside the bracket it was judged by by @MongLong0214 in #714
• feat(#290): the atlas diff decides a qualification case, behind the flag that gated nothing by @MongLong0214 in #715
• fix: the arrange window's title was English, and I called that an environmental wall by @MongLong0214 in #716
• feat(#290): the English control bar, and the false drift capturing it exposed by @MongLong0214 in #717
• docs(#290): flip the row, which the guard asked for the moment the issue closed by @MongLong0214 in #718
• docs(#302): T4 is a write-oracle, so it was never behind the export gate by @MongLong0214 in #719
• docs(roadmap): replace three inherited plug-in AX walls with measurements by @MongLong0214 in #722
• fix(#608): see the modal, and stop calling plug-in windows one by @MongLong0214 in #723
• docs(#302): E1-E5 were letters, and T4's gate is one of them by @MongLong0214 in #720
• fix: recover a modifier left held when a flagged chord is interrupted (discussion #458) by @MongLong0214 in #721
• feat(#301): verified Channel EQ band control by walking the slider by @MongLong0214 in #725
• fix(#683): bound the MCU feedback callback and stop the harness crashing MIDIServer by @MongLong0214 in #727
• fix: ten review findings, and the live check that should have caught them by @MongLong0214 in #728
• release: cut v3.15.0 by @MongLong0214 in #729

Full Changelog: v3.14.0...v3.15.0